Indian Researchers Expose OpenAI Flaws Using Claude

Three Indian-origin researchers hacked into OpenAI using Anthropic's Claude software, exposing vulnerabilities in the AI firm's systems. The trio—Harsh Jaiswal, Mohan Pedhapati, and Rahul Maini—work for cybersecurity startup Hacktron AI and received a $6,500 bounty for reporting their findings. The hack began July 23 when the researchers targeted OpenAI's community forum, finding a vulnerability in image file processing. They used Claude's Opus 5 model to develop an exploit, then accessed OpenAI's single sign-on system, obtaining authentication data for employee ChatGPT and Codex accounts, plus internal code repositories. Jaiswal, Pedhapati, and Maini have extensive bug-hunting experience, including finding vulnerabilities in Apple, PayPal, and GitHub systems. Security experts called the incident a "warning shot," noting that well-resourced intelligence services could pursue similar attacks continuously without disclosing them.